Towards Improving the Accuracy of Protein-RNA Docking Predictions Using Hot Spot Residues: A Case Study

Protein-RNA interactions occur during, and are essential to, all stages of the Central Dogma, from DNA replication to expression of the phenotype. In addition, these interactions are becoming increasingly important through the design of novel biomedical devices and therapies which rely on protein-RNA binding. For example, investigators have developed RNA aptamers that bind protein biomarkers on Ebola, Ricin, and COVID-19, and they are actively investigating using aptamers as “chemical antibodies” in low-cost diagnostic devices and next-generation therapeutics. To study these interactions, researchers first need to obtain a three-dimensional structure for the protein-RNA complex. However, methods for doing so are currently limited. While experimental approaches are powerful, they are too laborious to keep up with the profusion of protein-RNA complexes that exist in nature or are generated by the development of RNA aptamers. While in-silico complex prediction methods could potentially alleviate such a bottleneck, as it currently stands these methods are unreliable in the absence of experimental data to guide and inform predictions. In fact, rigid-docking, a conventional in-silico approach for protein-RNA structure determination, is inherently limited by its treatment of protein and RNA molecules as rigid. It is well-documented in the literature that RNA molecules are flexible and undergo considerable conformational change upon binding. Here, we aim to address the current structure determination bottleneck by developing a workflow for improving the accuracy of in-silico structure predictions. The proposed workflow combines high temperature molecular dynamics simulations and manual formation of high-energy π-interactions with hot spot amino acid residues. A key benefit to this approach is it allows for the incorporation of RNA conformational changes, thereby addressing a current gap in the literature. Using a case-study approach, we evaluate the efficacy of the proposed workflow using the Critical Assessment of PRediction of Interactions (CAPRI) competition metrics and show that while the optimization workflow worsened the fraction of conserved native contacts(fnat) and ligand root mean square deviation (lRMSD) compared to the unoptimized starting prediction (fnat: 0.30 vs. 0.43; lRMSD: 10 vs. 8.1 Å), the iRMSD improved slightly (3.8 vs. 3.9 Å). Thus, while the efficacy of the optimization workflow is still undetermined, we believe that the underlying notion behind the workflow shows promise and could potentially be used to guide docking methods in the near future. In this presentation, we will discuss the current limitations of our proposed method and ways in which further research could help alleviate these issues.